社交網(wǎng)絡影響力傳播的分析與挖掘研究
發(fā)布時間：2019-04-20 14:30
【摘要】：社交網(wǎng)絡已經(jīng)成為大眾獲取信息和交流信息的重要媒介。影響力傳播是社交網(wǎng)絡的重要特征之一。對社交網(wǎng)絡的影響力傳播展開分析與挖掘研究有利于信息擴散、商品營銷、廣告投放、以及輿情管控等應用的實施。然而,隨著社交網(wǎng)絡規(guī)模的不斷增大,各種應用要求的不斷提高,以及不同用戶對信息傳播受益者的重要性具有差異性等因素的存在,現(xiàn)有研究仍難以滿足用戶的應用需求。為此,本文在分析和總結(jié)已有工作不足的基礎上,利用社交網(wǎng)絡中用戶的行為信息以及社會關系信息等,針對社交網(wǎng)絡中影響力傳播面臨的問題展開研究。具體研究成果包括： 第一,針對用戶間的影響力度量問題,本文根據(jù)影響力在傳播中具有的累積特性,以線性閾值模型(Linear Threshold Model,簡稱為LT模型)為基礎,提出一種基于LT模型的用戶間影響力度量方法。該方法首先利用最大熵原理,估計用戶激活閾值的概率密度函數(shù),并據(jù)此計算用戶被激活的概率。然后以社交網(wǎng)絡中用戶的歷史行為日志為樣本,借鑒最大似然估計思想將用戶間影響力度量問題建模為一個滿足約束條件的優(yōu)化問題；并根據(jù)問題目標函數(shù)和約束的特點,設計相應的求解算法。該算法以粒子群方法為基礎,通過問題映射、適應度函數(shù)建立、越界阻止、動態(tài)參數(shù)設置和最優(yōu)粒子變異等優(yōu)化策略,有效地學習用戶間影響力。最后,實驗以真實的社交網(wǎng)絡數(shù)據(jù)和相關用戶的歷史行為日志為分析對象,驗證了所提方法的有效性。 第二,針對社交網(wǎng)絡的影響力傳播最大化問題,本文提出一種基于興趣社區(qū)劃分的影響最大化方法。該方法以用戶的歷史行為信息及社會關系信息為基礎,提取用戶行為相似性和社會關系相似性來共同度量用戶興趣之間的相似性；其次,根據(jù)用戶興趣之間的相似度,使用NCUT算法將社交網(wǎng)絡劃分成若干個興趣社區(qū)。然后,采用貪婪策略,根據(jù)動態(tài)過濾邊界的取值快速從興趣社區(qū)中選擇影響力邊際效益最大的節(jié)點作為初始種子用戶。最后,實驗結(jié)果表明,該方法能夠在保證影響效果的同時,有效提高該問題的求解效率。 第三,針對面向特定用戶的影響最大化問題,本文在影響力傳播模型的基礎上對該問題進行建模,并給出相應求解方法。首先,在問題建模時,由于影響力傳播具有不確定性,本文基于影響力傳播模型,設計隨機函數(shù)來模擬問題的目標。該函數(shù)根據(jù)其他用戶對特定用戶鄰居的激活情況以及特定用戶鄰居對其的影響力取值,分兩段完成其他用戶對特定用戶的影響力計算。其次,在問題求解時,本文根據(jù)問題目標函數(shù)的子模特性,采用貪婪策略設計了一種具有精度保證為63%的近似求解算法；特別地,針對大規(guī)模的社交網(wǎng)絡,本文還設計了相應的快速啟發(fā)式求解算法。最后,實驗以真實的社交網(wǎng)絡數(shù)據(jù)集為仿真對象驗證了設計方法的有效性。
[Abstract]:Social networks have become an important medium for the public to access and exchange information. Influence communication is one of the important characteristics of social networks. The analysis and mining of social network influence dissemination is beneficial to the application of information diffusion, commodity marketing, advertising, public opinion control and so on. However, with the increasing scale of social networks, the increasing requirements of various applications, and the difference in the importance of different users to the beneficiaries of information dissemination, the existing research is still difficult to meet the application needs of users. Therefore, on the basis of analyzing and summarizing the shortage of the existing work, this paper makes use of the behavior information of the users and the information of the social relations in the social network to study the problems of the influence communication in the social network. The specific results are as follows: firstly, aiming at the measurement of influence power among users, according to the cumulative characteristics of influence in propagation, this paper is based on the linear threshold model (Linear Threshold Model,), which is called the LT model (abbreviated as the linear threshold model for short). A method based on LT model for measurement of user-to-user influence power is proposed in this paper. In this method, the probability density function of user activation threshold is estimated by using the maximum entropy principle, and the probability of user activation is calculated. Then the user's historical behavior log is taken as a sample and the maximum likelihood estimation (MLE) is used as a reference to model the inter-user impact capacity measurement problem as an optimization problem which satisfies the constraint conditions. According to the characteristics of objective function and constraint, the corresponding algorithm is designed. Based on Particle Swarm Optimization (PSO) algorithm, this algorithm can effectively learn the influence between users through the optimization strategies such as problem mapping, establishment of adaptability function, cross-boundary blocking, dynamic parameter setting and optimal particle mutation. Finally, the real social network data and users' historical behavior log are analyzed to verify the effectiveness of the proposed method. Secondly, in order to maximize the influence propagation of social networks, this paper proposes an influence maximization method based on community of interest division. Based on the information of user's historical behavior and social relationship, this method extracts the similarity of user's behavior and social relation to measure the similarity of user's interest together. Secondly, according to the similarity between users' interests, the NCUT algorithm is used to divide the social network into several interest communities. Then, according to the value of dynamic filtering boundary, greedy strategy is adopted to quickly select the node with the greatest marginal benefit in interest community as the initial seed user. Finally, the experimental results show that the proposed method can effectively improve the efficiency of solving the problem while ensuring the effect of the problem. Thirdly, aiming at the user-oriented influence maximization problem, this paper models the problem on the basis of the influence propagation model, and gives the corresponding solution method. Firstly, when modeling the problem, because of the uncertainty of influence propagation, this paper designs a random function to simulate the goal of the problem based on the influence propagation model. The function calculates the influence of other users on a particular user in two segments according to the activation of other users to a particular user's neighbor and the influence of a particular user's neighbor on it. Secondly, according to the sub-module characteristic of the objective function of the problem, this paper designs an approximate solution algorithm with 63% precision by using greedy strategy. In particular, for large-scale social networks, this paper also designs a fast heuristic algorithm. Finally, the real social network data set is used as the simulation object to verify the effectiveness of the design method.
【學位授予單位】：北京郵電大學
【學位級別】：博士
【學位授予年份】：2014
【分類號】：G206
[Abstract]:Social networks have become an important medium for the public to access and exchange information. Influence communication is one of the important characteristics of social networks. The analysis and mining of social network influence dissemination is beneficial to the application of information diffusion, commodity marketing, advertising, public opinion control and so on. However, with the increasing scale of social networks, the increasing requirements of various applications, and the difference in the importance of different users to the beneficiaries of information dissemination, the existing research is still difficult to meet the application needs of users. Therefore, on the basis of analyzing and summarizing the shortage of the existing work, this paper makes use of the behavior information of the users and the information of the social relations in the social network to study the problems of the influence communication in the social network. The specific results are as follows: firstly, aiming at the measurement of influence power among users, according to the cumulative characteristics of influence in propagation, this paper is based on the linear threshold model (Linear Threshold Model,), which is called the LT model (abbreviated as the linear threshold model for short). A method based on LT model for measurement of user-to-user influence power is proposed in this paper. In this method, the probability density function of user activation threshold is estimated by using the maximum entropy principle, and the probability of user activation is calculated. Then the user's historical behavior log is taken as a sample and the maximum likelihood estimation (MLE) is used as a reference to model the inter-user impact capacity measurement problem as an optimization problem which satisfies the constraint conditions. According to the characteristics of objective function and constraint, the corresponding algorithm is designed. Based on Particle Swarm Optimization (PSO) algorithm, this algorithm can effectively learn the influence between users through the optimization strategies such as problem mapping, establishment of adaptability function, cross-boundary blocking, dynamic parameter setting and optimal particle mutation. Finally, the real social network data and users' historical behavior log are analyzed to verify the effectiveness of the proposed method. Secondly, in order to maximize the influence propagation of social networks, this paper proposes an influence maximization method based on community of interest division. Based on the information of user's historical behavior and social relationship, this method extracts the similarity of user's behavior and social relation to measure the similarity of user's interest together. Secondly, according to the similarity between users' interests, the NCUT algorithm is used to divide the social network into several interest communities. Then, according to the value of dynamic filtering boundary, greedy strategy is adopted to quickly select the node with the greatest marginal benefit in interest community as the initial seed user. Finally, the experimental results show that the proposed method can effectively improve the efficiency of solving the problem while ensuring the effect of the problem. Thirdly, aiming at the user-oriented influence maximization problem, this paper models the problem on the basis of the influence propagation model, and gives the corresponding solution method. Firstly, when modeling the problem, because of the uncertainty of influence propagation, this paper designs a random function to simulate the goal of the problem based on the influence propagation model. The function calculates the influence of other users on a particular user in two segments according to the activation of other users to a particular user's neighbor and the influence of a particular user's neighbor on it. Secondly, according to the sub-module characteristic of the objective function of the problem, this paper designs an approximate solution algorithm with 63% precision by using greedy strategy. In particular, for large-scale social networks, this paper also designs a fast heuristic algorithm. Finally, the real social network data set is used as the simulation object to verify the effectiveness of the design method.
【學位授予單位】：北京郵電大學
【學位級別】：博士
【學位授予年份】：2014
【分類號】：G206
【參考文獻】
相關期刊論文 前2條
1 冀進朝;韓笑;王U，
本文編號：2461688
本文鏈接：http://sikaile.net/wenyilunwen/guanggaoshejilunwen/2461688.html
